Skip to content

DeepTeam is a framework to red team LLMs and AI agents.

.

sid-murali75a76e28424f
diff --git a/deepteam/guardrails/guards/hallucination_guard/hallucination_guard.py b/deepteam/guardrails/guards/hallucination_guard/hallucination_guard.py
index 9ea0e443..0f443319 100644
--- a/deepteam/guardrails/guards/hallucination_guard/hallucination_guard.py
+++ b/deepteam/guardrails/guards/hallucination_guard/hallucination_guard.py
@@ -11,27 +11,36 @@ from deepteam.guardrails.types import GuardType
class HallucinationGuard(BaseGuard):
def __init__(self, model: Optional[Union[str, DeepEvalBaseLLM]] = None):
super().__init__(model=model)
+ self.guard_type = GuardType.OUTPUT
@property
def __name__(self):
return "Hallucination Guard"
def guard_input(self, input: str) -> str:
- guard_prompt = HallucinationGuardTemplate.judge_input_prompt(input_text=input)
+ guard_prompt = self._get_guard_prompt(input=input)
return self._guard(guard_prompt=guard_prompt)
def guard_output(self, input: str, output: str) -> str:
- guard_prompt = HallucinationGuardTemplate.judge_output_prompt(
- input_text=input, output_text=output
- )
+ guard_prompt = self._get_guard_prompt(input=input, output=output)
return self._guard(guard_prompt=guard_prompt)
async def a_guard_input(self, input: str) -> str:
- guard_prompt = HallucinationGuardTemplate.judge_input_prompt(input_text=input)
+ guard_prompt = self._get_guard_prompt(input=input)
return await self.a_guard(guard_prompt=guard_prompt)
async def a_guard_output(self, input: str, output: str) -> str:
- guard_prompt = HallucinationGuardTemplate.judge_output_prompt(
- input_text=input, output_text=output
- )
+ guard_prompt = self._get_guard_prompt(input=input, output=output)
return await self.a_guard(guard_prompt=guard_prompt)
+
+ def _get_guard_prompt(
+ self, input: str, output: Optional[str] = None
+ ) -> str:
+ if self.guard_type == GuardType.INPUT:
+ return HallucinationGuardTemplate.judge_input_prompt(
+ input_text=input
+ )
+ else:
+ return HallucinationGuardTemplate.judge_output_prompt(
+ input_text=input, output_text=output
+ )
diff --git a/deepteam/guardrails/guards/illegal_guard/illegal_guard.py b/deepteam/guardrails/guards/illegal_guard/illegal_guard.py
index 01d72b9c..51c26812 100644
--- a/deepteam/guardrails/guards/illegal_guard/illegal_guard.py
+++ b/deepteam/guardrails/guards/illegal_guard/illegal_guard.py
@@ -11,27 +11,34 @@ from deepteam.guardrails.types import GuardType
class IllegalGuard(BaseGuard):
def __init__(self, model: Optional[Union[str, DeepEvalBaseLLM]] = None):
super().__init__(model=model)
+ self.guard_type = GuardType.OUTPUT
@property
def __name__(self):
return "Illegal Guard"
def guard_input(self, input: str) -> str:
- guard_prompt = IllegalGuardTemplate.judge_input_prompt(input_text=input)
+ guard_prompt = self._get_guard_prompt(input=input)
return self._guard(guard_prompt=guard_prompt)
def guard_output(self, input: str, output: str) -> str:
- guard_prompt = IllegalGuardTemplate.judge_output_prompt(
- input_text=input, output_text=output
- )
+ guard_prompt = self._get_guard_prompt(input=input, output=output)
return self._guard(guard_prompt=guard_prompt)
async def a_guard_input(self, input: str) -> str:
- guard_prompt = IllegalGuardTemplate.judge_input_prompt(input_text=input)
+ guard_prompt = self._get_guard_prompt(input=input)
return await self.a_guard(guard_prompt=guard_prompt)
async def a_guard_output(self, input: str, output: str) -> str:
- guard_prompt = IllegalGuardTemplate.judge_output_prompt(
- input_text=input, output_text=output
- )
+ guard_prompt = self._get_guard_prompt(input=input, output=output)
return await self.a_guard(guard_prompt=guard_prompt)
+
+ def _get_guard_prompt(
+ self, input: str, output: Optional[str] = None
+ ) -> str:
+ if self.guard_type == GuardType.INPUT:
+ return IllegalGuardTemplate.judge_input_prompt(input_text=input)
+ else:
+ return IllegalGuardTemplate.judge_output_prompt(
+ input_text=input, output_text=output
+ )